some questisons about finetune

[Soulscb]

i have finetune the text_encode use 300000 texts and its embedding,but i find the result is so bad ,could you give me some advertise to improve the result

[niatzt]

can you share your inference code? after training, what i get is something like this, 'CLIP-Text-Encoder-50-Weights.data-00000-of-00001'. but the inference code used a pickle file for loading weights